Check whether a planned feature already exists with parallel research agents

A solo developer sent 13 research agents plus a refuting pass to size a market and found the planned feature was already free in Claude Desktop and the ChatGPT app.

Done withCustom agent· Oct 26, 2026

What they did

jidonglab is building Refmade, a desktop app that reads Claude Code and Codex session logs and draws them as a pixel office, with coding agents writing most of the code. On 27 September 2026 jidonglab wrote down the next wanted feature, a window where people who dislike the terminal can type a task for Claude Code or Codex and watch it run, and asked for market size, viability and target users. Thirteen research agents worked the question, and a second pass tried to refute what they found. The output was a memo citing about 60 sources, where every number carried one of six labels: measured, announced, reported, estimated, assumed or proposed.

How it went

The agents agreed the feature was already included in apps people pay for: the Code tab in Claude Desktop, background sessions in claude agents, and Codex inside the ChatGPT desktop app. Free open-source tools such as Pixel Agents covered parts of the rest. The only gap left was one screen showing both vendors' sessions from logs alone. The memo advised a read-only dashboard first; the author overrode that and kept the send-work window, knowing it would not set the app apart. The labels showed the market-size estimate rested on unmeasured assumptions, so its high and low ends were tens of times apart.

Worth knowing

The post does not say which tool ran the research agents. The author rechecked the key claims against vendor pages before publishing, and the memo itself warned that not finding a competitor is not proof none exists. The author found the evidence of user pain thin: a few posts by developers who had already solved it with free tools.

Try it

With Custom agent· 2 parts to fill in

I am thinking of building [feature] for [product]. Research whether it is worth building: market size, existing tools that already do it (free and paid), who would use it, and the gap left over. Split the work across several research passes, then run a separate pass whose only job is to refute the findings. Label every number as measured, announced, reported, estimated, assumed or proposed, and cite a source for each claim. Stop when you have a memo with a comparison table of existing tools and a recommendation for the smallest first version; do not start building anything.

Square brackets are the parts to fill in. The prompt is ours; the story is theirs.

Where they said it

Related